Drivers of Consumer Resistance to AI

Product/Service InnovationZehnle, M.; Hildebrand, C.; Valenzuela, A. · 2025International Journal of Research in Marketing
Topicsartificial intelligence·ai aversion·meta-analysis·generative ai·algorithms·consumer response·anthropomorphism

Your leadership team wants one company-wide assumption about whether customers accept AI. Resistance here is the gap between how consumers respond to an AI and to a human doing the same task, negative when they respond worse to AI. Which AI services can scale now, and which need more evidence?

Consumer resistance to AI varies sharply by type, autonomy and use case.

Data chart

Consumer resistance is highest for embodied AI robots

Gap between consumer responses to an AI and to a human doing the same task, in standard deviations; below zero means consumers responded worse to the AI (0.2 is small, 0.8 is large).

No difference from a humanEmbodied AI robots-0.83AI algorithms-0.24AI assistants-0.22AI systems-0.17
  • Robots were associated with the largest negative response gap.
  • These averages do not isolate the effect of renaming a service.

Embodied AI robots were associated with the largest negative response gap.

Averages from experiments comparing AI with a human doing the same task; robot studies account for 13 of 440 results. Robot studies mainly concern high-stakes settings; this comparison does not isolate the effect of changing a service's name.

Data chart

Resistance has narrowed but not disappeared

Average gap between consumer responses to an AI and to a human doing the same task, for studies in each period, in standard deviations; below zero means consumers responded worse to the AI.

No difference from a humanBefore 2010-0.33From 2010 to 2014-0.47From 2015 to 2019-0.42From 2020 to 2025-0.12
  • The latest studies show the smallest gap, still below zero.
  • Year by year, the gap neared zero around 2021, then widened.

Studies from 2020 to 2025 show a much smaller gap than studies from any earlier period, but consumers still respond worse to the AI than to a human.

Averages for studies in four periods of AI development, from 14, 29, 97 and 300 results; observed across publication periods, not a forecast of when or whether resistance will disappear.

Action guide

  1. Test AI service names before approving positioning.Compare names for the same service; the studies did not isolate naming effects.
  2. Approve separate rollout plans by use case.Require validation before scaling services in transportation, legal and public safety or social welfare, or high-stakes uses such as diagnosis. Let operational, lifestyle and basic healthcare support uses move faster.
  3. Set a portfolio rule for visible AI autonomy.Test each use case before presenting your AI as deciding on its own.
  4. Show customers clear evidence of how well each AI service performs.Consumers strongly resist AI described as worse than a human, but show no reliable resistance to AI described as better.
  5. Require behavioral evidence from a current pilot before approving scale.Treat the overall average as a starting point only. Never shelve a use case on survey resistance alone.

Evidence

  • Resistance averages -0.21 standard deviations, where 0.2 is small and 0.8 large. A new study could find anything from -1.20 to 0.78.
  • Robots, AI with a physical body, averaged -0.83, against -0.17 to -0.24 for AI called algorithms, assistants or systems, but robot studies mostly concern high-stakes settings.
  • Resistance ranges from -0.80 in transportation to -0.12 in operations and management, with legal and public safety and social welfare also high. In healthcare, AI diagnosis averaged -0.26, while basic support such as patient onboarding showed no reliable resistance.
  • AI with extensive autonomy averaged -0.28, against -0.05 with limited autonomy. AI described as worse than a human averaged -0.53, while AI described as better showed no reliable resistance.
  • Studies from 2020-2025 found a smaller average gap than studies from 2015-2019 (-0.12 against -0.42), though consumers still favored humans.
  • Studies showing each person only the AI or only a human found a smaller gap than side-by-side comparisons (-0.19 against -0.38). Behavioral measures found less resistance than self-reports (-0.14 against -0.23), possibly by chance; field studies found no less.

Key takeaway

Consumer resistance to AI is small on average but varies sharply by type, autonomy and use case.

Limitations

The studies are experiments, mostly with North American samples, in which consumers know whether they face AI or a human. They do not measure adoption, revenue or profit.

Source

Zehnle, M., Hildebrand, C., & Valenzuela, A. (2025). Not all AI is created equal: A meta-analysis revealing drivers of AI resistance across markets, methods, and time. International Journal of Research in Marketing. https://doi.org/10.1016/j.ijresmar.2025.02.005

https://doi.org/10.1016/j.ijresmar.2025.02.005